MongoDB has announced new capabilities that give AI agents direct access to live operational data, alongside high-precision retrieval built directly into MongoDB Atlas, unveiled at the company’s MongoDB.local Build Fest.

The MongoDB Atlas Managed MCP Server is now available, offering a fully hosted way to connect AI agents to Atlas without additional infrastructure. Developers can add Atlas to coding tools including Claude Code, Codex, Grok Build and Devin, letting coding agents query, inspect and update live operational data as they work.

Voyage AI retrieval built into Atlas

MongoDB is also folding Voyage AI’s retrieval capabilities directly into Atlas. Automated Embeddings now automatically embed new and updated data to help AI applications retrieve current context, while a new Atlas Embedding and Reranking API gives applications direct access to MongoDB’s embedding and reranking models.

• A new voyage-code-4 model purpose-built for coding agents, matching information to code with higher precision and lower cost than earlier models
• Vector search in Atlas Stream Processing, extending retrieval to data in motion
• Customers including the Financial Times and Eve are already using the capabilities to improve retrieval accuracy while managing costs

Betting on the agentic coding stack

The announcements point to MongoDB’s push to embed itself deeper into the agentic coding stack, as developers increasingly rely on AI agents that need live, accurate access to operational data rather than static snapshots.

By hosting the MCP server directly and building retrieval into Atlas rather than requiring separate infrastructure, MongoDB is betting that lowering the operational overhead for developers will drive faster adoption of agent-based workflows.